Feature extraction in acoustic signals using the BCM learning rule

We apply the Bienenstock, Cooper, and Munro (1982) theory of visual cortical plasticity to the problem of extracting features (i.e., reduction of dimensionality) from acoustic signals; in this case, labeled samples of marine mammal sounds. We first implemented BCM learning in a single neuron model, trained the neuron on samples of acoustic data, and then observed the response when the neuron was tested on different classes of acoustic signals. Next, a multiple neuron network was constructed, with lateral inhibition among the neurons. By training neurons to be selective to inherent features in these signals, we are able to develop networks which can then be used in the design of an automated acoustic signal classifier.